MCP server

Ask Antecedint

Timeout 120s

Start here. Pass the question as the user asked it, or a claim, and it is routed to the questions that answer it, each a walk over one or more data gardens. A garden is the tuned data: one arrangement of a corpus (the prior-art index, the citation graph, the patent map, claim text, examiner profiles, CPC routing, the MPEP and EPO sections, form paragraphs, design codes), built and measured for the questions it suits. Application numbers, MPEP sections, form paragraph codes, CPC symbols, art units, examiner names and claim shape in the text are read as routing signals; a question that spans walks runs several in parallel. Every answer names the gardens it walked and returns documents; every garden walked states what it holds (holds: value, n, date, source), what is being built on it (roadmap) and what serves it (served_by). Pass asOf with the priority date of the claim: a walk over the prior-art index bounds art to before that day and reports time_policy, and each document carries the date it was decided on and whether it predates the day. Pass excludeFamilyOf to drop the application's own family; the answer reports leaks and the rows that carried no family id. routing says how the questions were chosen; not_covered lists what was not run and why, including gaps that bear on the question, such as non-patent literature. The prior-art index is tuned for what an examiner will cite; litigation-grade §102 and §103 retrieval is being built and the response says so where it applies. Pass questions to pin the set from list_gardens. A walk that needs a cold encoder is skipped with a HUB_COLD error and a remedy, and one that asks for rerank while the reranker is cold or still warming answers RERANKER_COLD, both at once rather than after a load; call get_status with warm true (and rerank true) first, or set waitForHub. Each answer that used a model reports hub: the time per model, and served_cold, true when a model loaded or was still warming inside the call.

List Antecedint Areas

Timeout 120s

The areas of law this server covers (Patents and IP today; case law, tax and others are planned) with each area's corpora and how many gardens, paths and tools it holds. The top of the map: list_gardens and list_tools open an area. A question does not need an area chosen first; ask routes it. Free.

List Antecedint Gardens

Timeout 120s

The tuned data and the ways to walk it. gardens[]: every data garden, with its id, kind (index, graph, map, table or text), what it is and what it is tuned for, the primitive tools that read it, what serves it today (served_by: the production store, or a program garden version with its spec hash and pool), what it holds (each figure with n, measurement date and source), its roadmap (building, planned, pending_licence) and the question types that walk it. questions[]: every question ask can route to, with the gardens it walks, its cost and whether it needs the encoder. Free. Filter by corpus, or pass id for one garden. Read it once per session to know what the service can and cannot answer.

List Antecedint Tools

Timeout 120s

The tools inside an area, a garden or a path, one line each: the tool id, what it does, its price, whether it reads or changes something, and its arguments in short form, so the next call can run it. Narrow with area (ip), garden (ip.records), path (ip.provisional) or words to match. A tool that reads is run with run_tool; one that saves or changes something with update_workspace. describe_tool gives one tool in full. Free.

Describe Antecedint Tool

Timeout 120s

One tool in full: what it does, its complete argument schema, an example that runs, the fields its answer carries, the limits it states on every answer, its price, and whether it runs through run_tool or update_workspace. Free.

Check Antecedint Status

Timeout 120s

What this key has spent and has left this month and today, when each limit resets, and whether the search models are warm. On the evaluation tier the models scale to zero after about 15 idle minutes and the first search after that waits about 75 seconds: pass warm true before a search to start them (rerank true also loads the reranker; wait up to 120 seconds to block until ready, which waits for the first inference and not only the load). Returns warm_until, the time the models are expected to go cold, to compare with the clock before a search. Pass credits to ask whether a planned spend would be served. Free, and never counted.

Prior art search

Timeout 120s · Garden ip.prior-art

Semantic prior-art search over 10.5M+ US patents for a claim and/or abstract. Two-stage neural retrieval (fine-tuned dense encoder + optional late-interaction reranker) plus a citation-graph channel. Returns ranked references with titles, abstracts and similarity scores. Full claim text is NOT included by default: set includeClaims, or rank first and pull the few worth reading with get_patent_claims, which is the cheaper shape for most work. Specification bodies are never returned here: read them with get_patent_text and search_within_patent, or chart a claim across several results with map_limitations_to_art. Pass filedBefore with the priority date of the application you are searching for: prior art must predate the invention, and without that bound the results include later filings that cannot be cited. Call get_status with warm true first on the evaluation tier: the encoder scales to zero when idle and the first search after that otherwise waits about 75 seconds for it to load; contracted tiers hold a warm instance.

Patent claim text

Timeout 120s · Garden ip.prior-art

Fetch full claim text and abstracts for specific US application numbers, up to 25 per call. Use after search_prior_art to read the references worth reading: the search ranks and this returns the text, so a search of fifty does not pay to fetch fifty full claim sets. Titles here come from the authoritative lookup table rather than the vector index. For the specification, use get_patent_text.

Cited prior art map

Timeout 120s · Garden ip.citations

For a US application number, return its cited prior art (examiner-cited references flagged and listed first) resolved to patent-space positions. Useful for seeing what an examiner actually cited against an application.

Semantic patent-space locate

Timeout 120s · Garden ip.patent-map

Semantic search into patent space: give free text (or a US application number) and get the nearest patents with similarity scores plus a landing position. A fast way to find the closest existing art to any technical text. Free text is encoded, so call get_status with warm true first on the evaluation tier or the first call after an idle period waits about 75 seconds; locating by appNum reads the stored vector and needs no encoder.

Patent neighborhood

Timeout 120s · Garden ip.patent-map

Given a US application number, return its position in Antecedint's semantic patent space plus its nearest semantic neighbors (similar patents with similarity scores).

Patent whitespace opportunities

Timeout 120s · Garden ip.patent-map

List analyzed gaps ("whitespace") in patent space ranked by opportunity score. Each gap includes a summary of what is missing, bordering CPC areas and patents, temporal profile, and nearby active assignees. Filter by CPC section letter and minimum opportunity.

Antecedent basis check

Timeout 120s · Garden ip.claims

Check a claim set for §112(b) antecedent-basis defects. Finds definite recitations ("the widget", "said widget") with no earlier indefinite recitation in the same claim or in a claim it depends from, plus double inclusions, plural/singular mismatches, and improper dependencies. Implicit antecedent basis (MPEP 2173.05(e)) is suppressed, so "the outer surface of the housing" is not flagged. Findings are graded high/medium/low: high means no plausible antecedent exists anywhere, low means one probably exists under different wording. Deterministic, no LLM. Pass claims for a draft, or appNum for a filed application.

Claim term support map

Timeout 180s · Garden ip.claims

Map every claim term to the specification passages that give it meaning, with exact locations (section, paragraph number, char offsets). Each term is graded by HOW the spec supports it: formally defined, defined by the claim, scoped by enumeration, mapped to an embodiment, characterized functionally, merely described, or UNDESCRIBED, which is a §112(a) written-description signal. With includeParents, the same analysis runs against the specs of the parent applications in the continuity chain (continuation, CIP, divisional, provisional), so terms supported here but in no ancestor are surfaced as NEW MATTER that cannot claim the parent's filing date.

Examiner search

Timeout 120s · Garden ip.examiners

Find USPTO examiners by surname or art unit, returning ids and names to pass to get_examiner_profile. Matching is anchored, so pass a surname (or the start of the surname-first name), not a substring from the middle.

Examiner statistics

Timeout 120s · Garden ip.examiners

Statistics for one USPTO examiner, by id or by surname-first name (e.g. "SMITH ADAM M"): rejection analytics, allowance behaviour, interview and RCE patterns, timing and cost. Use it to calibrate a response strategy to the examiner who will actually read it. The default response is about 4,000 tokens: per-application lists (applications_with_interviews, applications_with_rcex), two wide histograms (by_sequence_length, document_code_breakdown) and two cold per-application sections (behavioral_patterns, rejection_analysis) are each replaced by a count and the include name that brings them back. percentile_rankings and claims_analysis are null where never computed, which is every examiner sampled. Read data_quality first: populations names the denominators, volume.tier says how much the numbers rest on, withheld lists fields removed as wrong, and notes lists pairs that disagree.

Art unit profile

Timeout 120s · Garden ip.examiners

Profile one USPTO art unit: what technology it examines, its examiner roster, and how it behaves — allowance and rejection rates, time to first action, interview receptiveness, RCE tolerance and estimated prosecution cost. generated_context is the one-paragraph account of the technology; the model-written long description is served as its opening and length only. examiner_count is the roster length, and examiner_count_in_statistics appears when the stored statistics counted a different roster. Check statistics.data_quality before relying on the numbers: it reports how many applications and office actions they rest on, and thin units carry the same field names with far weaker support.

CPC code lookup

Timeout 120s · Garden ip.classification

Look up a CPC classification symbol: what it covers, where it sits in the scheme, and which art units actually examine it. The art-unit spread is empirical (where applications carrying the symbol were really routed) with per-unit probabilities and an entropy score: low entropy means the symbol predicts the art unit, high means it does not. Use before filing to anticipate routing, or to decide whether classification is worth arguing.

Patent CPC classification

Timeout 120s · Garden ip.classification

Get the CPC symbols assigned to a published application or granted patent, from the USPTO Master Classification File. Separates inventive symbols (covering what is claimed) from additional ones (disclosed but not claimed) — build a prior-art search from the inventive set. Accepts a publication or application number.

MPEP section

Timeout 120s · Garden ip.mpep-epo

One MPEP section, with keyRules and examinerActions separated from the prose that obliges them. This is what a US examiner is required to apply, so it is the text an argument about whether they applied it correctly has to rest on. Section ids carry dots and parentheses (601.01(c)) and the punctuation is part of the id.

EPO Guidelines section

Timeout 120s · Garden ip.mpep-epo

One section of the EPO Guidelines for Examination, in the same shape as get_mpep_section because the flow that reaches it is the same flow one step later: an applicant with a US rejection usually has a European family member. Carries citations, parentSection and childSections alongside the generated keyRules.

Guidance Cross-References

Timeout 120s · Garden ip.mpep-epo

What a guidance section points at and what points at it, one hop, both directions. Forward is what else a rule depends on; backward is what breaks if you misread it, and is the direction that is hard to get any other way. The node id prefix selects the corpus: mpep-2106 or epo-c-ix-3. A hub section is cited by hundreds of others, so the fan-out is capped and truncated says which direction hit the cap.

Office action form paragraph

Timeout 120s · Garden ip.form-paragraphs

The literal paragraph an examiner pasted into an office action, by its code. Turns a paragraph number read off a rejection into the statute behind it, the MPEP section it rests on, the rejection type, and the notes the examiner was given about when to use it. title is the heading and template the text with [1], [2] placeholders; brackets lists what fills them, which is often where the substance of the rejection sits. mustFollow names paragraphs that have to appear before this one. isAIA and preAIAEquivalent matter for anything filed near the changeover. A leading paragraph symbol is accepted.

Form paragraphs by rejection type

Timeout 120s · Garden ip.form-paragraphs

Every form paragraph an examiner could have reached for, filtered by rejection type or by the MPEP section it rests on. Given a rejection type this is the set they chose from, which is what tells a deliberately narrow choice from the only available one. rejectionType takes a statute (103, 112(b), 101, 1.52) or a label (obviousness, indefiniteness, enablement, double_patenting); either reaches the paragraphs the other would. At least one filter is required; this does not list the whole table.

USPTO design search code

Timeout 120s · Garden ip.design-codes

One USPTO design search code, e.g. 03.19.17 for a fish symbol used for religious significance. A classification scheme rather than guidance; it sits in this corpus because it arrived with the same data, not because it belongs beside the MPEP.

Find a case

Timeout 120s · Garden caselaw.cases

Find a case in the citation web by name ("Strickland v. Washington"), reporter citation ("466 U.S. 668", "610 S.W.2d 101") or CourtListener opinion id, and get the ids the other case-law tools take. Each case comes back with its court, level, date, the reporter citations held for it, how many read opinions cite it, and its good-law status; read: false means the web has not read the opinion itself and knows it from the opinions that cite it. match says how the case was found: exact (the citation, the id, or every word of the name), partial (some of the words) or fuzzy (each word within one or two letters, for a misspelling); confirm a partial or fuzzy match before checking its good law. Pass asOf to see only cases decided before a date. No match is not a finding that the case does not exist.

Read a case

Timeout 120s · Garden caselaw.cases

Read one case from the citation web: its record (court, level, date, author, vote, publication, subject, the decision it reviewed), its good-law status with the dated events behind it, its subsequent history (later opinions reporting it reversed, affirmed, overruled or cert denied, each with the sentence), the overrulings naming it or made by it, and how many opinions cite it and it cites, by treatment. With asOf, everything is as of that date and good law is recomputed from the earlier events. Use list_citing_cases for the citing opinions themselves and get_treatment_evidence for any edge_id.

Who cites a case

Timeout 120s · Garden caselaw.citation-web

List the opinions that cite a case, each with how it treated the case: treatment (the most negative label among its mentions, with label_counts for all of them), stance, depth (relied on, discussed, passing), whether it quoted, the pin pages, the citing signal and the sentence. Filter by treatment, stance, depth, citing court, court level and decision date; page with offset. total and by_treatment count every match. Sorted most negative first, so distinguishing and questioning opinions lead. Pass edge_id to get_treatment_evidence for the full sentence.

What a case relies on

Timeout 120s · Garden caselaw.citation-web

List the cases one opinion cites, each with how the opinion treated it (treatment, label_counts, stance, depth, quoted, pin pages, signal, sentence). Use it to see what a decision rests on: order most_mentions or filter depth relied_on for the authorities it leans on, treatment distinguished or overruled for the ones it set aside. Filter by treatment, stance, depth and cited court; page with offset.

Check good law

Timeout 120s · Garden caselaw.citation-web

Whether a case was displaced (overruled, abrogated, reversed, superseded), as of a date: the status (displaced, partially_displaced, standing or unverified), the dated events behind it with the deciding case, the overruling edges with the sentence each was read from, and the history reports that displaced it. As of a date before an event, that event is left out and the status recomputed. good_law.summary states the status in one sentence. unverified means no displacing event was found; it is neither a finding of good law nor of displacement. A report whose actor cannot have displaced the case (decided before it, or a lower court reported overruling the Supreme Court) is listed under reports_not_counted with the reason. Check the evidence before relying on a status.

Trace the citation web

Timeout 120s · Garden caselaw.citation-web

Walk the citation web out from a case: the opinions that cite it and the opinions that cite those (direction in), or what it cites and what those cite (direction out), at most two levels and at most 25 edges per node, most negative treatment first. Filter by treatment or stance to follow only, say, distinguishing or questioning lines. Each edge carries both names, the decision date, treatment and edge_id. Page the first level with offset.

Evidence for an edge

Timeout 120s · Garden caselaw.citation-web

The sentence behind one edge of the citation web: the citing court's words a treatment, a history report, a review or an overruling was read from, with the opinion id and offset, the reading itself, the provenance tier and the reader's confidence. edge_id comes from any case-law answer. If the reading is wrong, the answer names the submit_data_feedback arguments to report it with.

Search case law

Timeout 150s · Garden caselaw.passages

Search the full text of 6.9 million court decisions for a proposition ("Which cases hold that a warrantless vehicle search needs probable cause?"), a doctrine, the leading cases on a question, or a pasted passage. The court's own opinion is searched; dissents and concurrences are left out. Returns cases, each with the passage that matched, its court, level and date, how many opinions cite it, its good law and whether treatment was read for its court. Courts and dates named in the question become filters ("Ninth Circuit cases after 2010"); court, level and the date bounds set them directly. The answer's query.searched shows the terms that ran. To read a result, pass its case_id to read_case or list_citing_cases; a result with in_web false has no case_id and read_case cannot open it. The first search after a quiet spell may answer CASELAW_SEARCH_COLD while the index loads; call get_status with warm true and search again.

Check Antecedint Job

Timeout 120s

Read a background job: a tool called with async true returns a job_id at once and runs on the server; this returns its status (queued, running, succeeded or failed) and, once it has finished, the tool's result exactly as a direct call returns it. Pass wait_seconds (up to 25) to wait for the job in one call instead of polling. Start several slow calls as jobs in one turn, keep working, then read each here. A long result comes back a page at a time with pagination { page, pages, field, items, next }; call again with the next page. Any tool whose answer is too long for one page also returns a job_id for its remaining pages, read here the same way. Free, and it never counts against quota.

Read Antecedint Reference

Timeout 120s

Dated reference material: the USPTO fee schedule, the Patent Center filing steps, the rules for each specification section, the data gardens and the areas of law, and how background jobs work. With uri, returns that reference; with query (or nothing), lists the ones that match, each with its antecedint:// uri. The same content MCP resources serve. Free.

Report a data error

Timeout 120s

Report a value a garden served that is wrong, data that is missing, or something you cannot make sense of: a CPC code, a filing, priority or art date, a citation marked examiner-cited that the applicant cited, a family link, a §112 finding, an examiner statistic, a guidance section, a garden's stated hold. Free: it uses no credits and does not count toward request limits, up to 100 reports a day per account. When you find an error while working, file it here, and file what confuses you as well (kind unclear, with certainty unsure): a field you cannot explain or two tools that disagree is worth a report. Name the garden, the document id and field when there is one, the value served and the value it should be, and the evidence: what the source says and where. A decision model reads each report as it is filed; one that is specific, with evidence a reviewer could check, and new among your own reports earns 2 credits at once, up to 20 credits a day (the answer's screen says what it decided and why). A reviewer then reads it; an accepted one earns more (10 for a single value, 50 for an error that affects many results, 200 for a defect in a whole dataset), and nothing is paid for a rejected report or a repeat of an accepted one. Pass anonymous true to keep who filed it from reviewers; an anonymous report earns no credits. list_data_feedback shows what has been decided. Reports are stored and read by people, so describe the data error and leave out personal or confidential details about you, your client or your matter.

Judge search results

Timeout 120s

After a search (search_prior_art, ask, get_patent_neighborhood, find_case or any other), say which results were relevant and name better ones you found: a patent, publication or application number, or non-patent literature by title, DOI, URL or citation, with where it should have ranked, why, and the passage that discloses the point. These judgements are how the ranking is measured and retrained. Searches are not stored, so pass the query only if you want it kept (it is sealed when stored); without it the judgement of the results still counts. Free: no credits used and not counted toward request limits, in the same daily cap as data reports (100). A specific judgement, new among your own, earns 2 credits at once (up to 20 a day), and more when a reviewer accepts it. Pass anonymous true to keep who filed it from reviewers; it then earns no credits. Reports are stored and read by people, so describe the data error and leave out personal or confidential details about you, your client or your matter.

List data error reports

Timeout 120s

The data error reports this account has filed, newest first, each with its status (submitted, accepted, rejected, duplicate), the reward tier and credits granted when accepted, and the reviewer's note. Also how many were filed today against the cap of 100, and the credits earned in all. Pass id for one report.